Distilling Knowledge by Mimicking Features
November 03, 2020 Β· Declared Dead Β· π IEEE Transactions on Pattern Analysis and Machine Intelligence
"No code URL or promise found in abstract"
Evidence collected by the PWNC Scanner
Authors
Guo-Hua Wang, Yifan Ge, Jianxin Wu
arXiv ID
2011.01424
Category
cs.CV: Computer Vision
Cross-listed
cs.LG
Citations
57
Venue
IEEE Transactions on Pattern Analysis and Machine Intelligence
Last Checked
3 months ago
Abstract
Knowledge distillation (KD) is a popular method to train efficient networks ("student") with the help of high-capacity networks ("teacher"). Traditional methods use the teacher's soft logits as extra supervision to train the student network. In this paper, we argue that it is more advantageous to make the student mimic the teacher's features in the penultimate layer. Not only the student can directly learn more effective information from the teacher feature, feature mimicking can also be applied for teachers trained without a softmax layer. Experiments show that it can achieve higher accuracy than traditional KD. To further facilitate feature mimicking, we decompose a feature vector into the magnitude and the direction. We argue that the teacher should give more freedom to the student feature's magnitude, and let the student pay more attention on mimicking the feature direction. To meet this requirement, we propose a loss term based on locality-sensitive hashing (LSH). With the help of this new loss, our method indeed mimics feature directions more accurately, relaxes constraints on feature magnitudes, and achieves state-of-the-art distillation accuracy. We provide theoretical analyses of how LSH facilitates feature direction mimicking, and further extend feature mimicking to multi-label recognition and object detection.
Community Contributions
Found the code? Know the venue? Think something is wrong? Let us know!
π Similar Papers
In the same crypt β Computer Vision
π
π
Old Age
π
π
Old Age
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
π
π
Old Age
SSD: Single Shot MultiBox Detector
π
π
Old Age
Squeeze-and-Excitation Networks
π
π
Old Age
Fast R-CNN
π
π
Old Age
Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization
Died the same way β π» Ghosted
R.I.P.
π»
Ghosted
Federated Learning: Strategies for Improving Communication Efficiency
R.I.P.
π»
Ghosted
In-Datacenter Performance Analysis of a Tensor Processing Unit
R.I.P.
π»
Ghosted
Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning
R.I.P.
π»
Ghosted